Virtualize SkGLContext with subclasses SkNativeGLContext and SkMesaGLContext, allow both in gm

Goal: Support multiple kinds of GL contexts within same build for various testing purposes. Mesa for consistent images, null context to cpu profile the gpu backend, validation context that checks for GL object leaks, etc.

Top level view of this change:
1) Adds some new GL functions to GrGLInterface needed by other parts of the change (finish, flush, queries).
2) Distinguishes between the "native" and "default" GrGLInterface. This allows us to create a native GL interface whether or it is not the default.
   The GrGLDefaultInterface_*.cpp files are now GrGLCreateNativeInterface_*.cpp
   Two new files GrGLDefaultInterface_none.cpp and GrGLDefaultInterface_native.cpp allow having no default (chrome) or using the native as the default (internal apps).
3) Virtualizes SkGLContext so that we can have multiple implementations in the same build.
   Currently two types: native, mesa. Future change: null gl context and validating gl contexts.
4) Enable osmesa in gm without requiring a rebuild (--mesagl switch)

All mesa code is protected by gyp var skia_mesa, since not all Mac and Linux machines have the necessary library. Tested gm on mac and linux with the --mesagl flag. The images look correct.

Once this is in I'd like to enable skia_mesa=1 on the mac and linux bots right away to keep it from bit rotting.

[bsalomon, 2011-10-19 18:56:41]

http://codereview.appspot.com/5307045/diff/1/src/gpu/GrGLProgram.h
File src/gpu/GrGLProgram.h (right):

http://codereview.appspot.com/5307045/diff/1/src/gpu/GrGLProgram.h#newcode44
src/gpu/GrGLProgram.h:44: k110_GLSLVersion,
The osmesa on either mac or linux (can't remember which) only supported GLSL
1.10. It turns out the shaders we were generating for 1.20 are valid 1.10 as
well.

[bungeman, 2011-10-19 20:23:06]

Other than the one comment, LGTM.

http://codereview.appspot.com/5307045/diff/1/bench/benchmain.cpp
File bench/benchmain.cpp (right):

http://codereview.appspot.com/5307045/diff/1/bench/benchmain.cpp#newcode508
bench/benchmain.cpp:508: timer.start();
Doesn't look like timer ever gets a GPU context?

[epoger, 2011-10-19 21:41:26]

gyp files LGTM

http://codereview.appspot.com/5307045/diff/3002/gyp/bench.gypi
File gyp/bench.gypi (left):

http://codereview.appspot.com/5307045/diff/3002/gyp/bench.gypi#oldcode65
gyp/bench.gypi:65: 'sources!': [
Casually wondering... When was this section ever active?  (what skia_os is not
one of win, mac, linux, freebsd, openbsd, solaris?)  iOS, I guess?

[bsalomon, 2011-10-20 13:28:47]

Closed with r's 2499, 2500, 2501, and finally 2505.

http://codereview.appspot.com/5307045/diff/3002/gyp/bench.gypi
File gyp/bench.gypi (left):

http://codereview.appspot.com/5307045/diff/3002/gyp/bench.gypi#oldcode65
gyp/bench.gypi:65: 'sources!': [
On 2011/10/19 21:41:26, epoger wrote:
> Casually wondering... When was this section ever active?  (what skia_os is not
> one of win, mac, linux, freebsd, openbsd, solaris?)  iOS, I guess?

I guess, but dunno.